InteractiveBuilt the same day the model shipped
Search by feeling, not filenames.
Type a mood, upload a photo, or click any result to explore similar vibes. Powered by Gemini Embedding 2, Google's first multimodal embedding model. Text and images share the same vector space, so your words match the visual content of photos.
Loading...
How it works
- Pre-embedded dataset: Photos were embedded (image pixels + caption text together) using
gemini-embedding-2-previewat 768 dimensions. The vectors are stored as a static JSON file. - Your query: When you type a vibe, we send your text to the same model to get a query embedding. Upload a photo and it gets embedded too, in the same vector space.
- Cosine similarity: Your query vector is compared against all photo vectors using cosine similarity (5 lines of math, runs in your browser). Top matches are shown.
- Click to explore: Every result is clickable. The app uses that image's existing embedding to find the most similar photos, no extra API call needed.
Photos from Unsplash.